PERSPECTIVES OF APPLICATION OF THE ADVANCED CLASSICAL ALGORITHM OF KOHONEN MAPS IN DISTRIBUTED ENERGY-CRITICAL SENSOR NETWORKS

Abstract

Topicality. The standard version of the Kohonen map algorithm can be improved by including the parallel finding of several winning neurons and the subsequent selection of regions of the sensory field that do not overlap. The mentioned approach, embedded in the methodology of modified Kohonen maps, is promising in relation to the directions of development of artificial intelligence based on artificial neural network structures. Compared to the standard version, modified Kohonen maps provide partial parallelization of object classification procedures within each of the identification iterations, both before the SNS training procedure and in the working mode. It is shown that a similar feature, but in a more developed form, is also characteristic of human intelligence and is realized in the neural structures of the human brain, which is the carrier of human intelligence. It is interesting that similar structures in human brain realize reduced energy consumption and increased survivability of the system with a limited fixed failure of a small number of elements. These qualities, apparently, can potentially be realized in the future during the further development of the modified Kohonen maps. The purpose of the work is to demonstrate the compliance of a number of features of the work of the MCC to the elements and features of the organization of information processes in the structures of human brain, which, in general, should be considered as a prototype in the development of artificial intelligence systems.